Material below summarizes the article Working Memory Replay Prioritizes Weakly Attended Events, published on August 14, 2017, in eNeuro and authored by Anna Jafarpour, Will Penny, Gareth Barnes, Robert T. Knight, and Emrah Duzel.
Working memory is a system with limited capacity that supports keeping information in mind for a short duration.
Two important questions are: One, what are the mechanisms by which information is retained during the gap between encoding and recall (that is, the maintenance period), and two, how is information within working memory prioritized?
These are often studied with a cuing procedure in which subjects are prompted to focus on one item. For example, if an event in a sequence of three events is cued, the cued event can be decoded during maintenance.
In this study, we investigated how the information about a sequence of multiple events is retained without cueing.
Our experiment was a modified version of the Sternberg task. In a trial, a sequence of three visual events had to be retained. Objects from three distinct visual categories (faces, manufactured objects, and natural events) were presented successively followed by a five-second delay period. After the delay, one question asked about an event’s identity and a second question queried the order of an event.
To decode the working memory content, we employed multivariate pattern analysis (MVPA) of magnetoencephalography (MEG) recording, which is a neuroimaging technique with temporal resolution of milliseconds. We previously showed content of memory is decodable using this technique.
Pattern classifiers were trained on categorical representations of visual stimuli during encoding and were used to decode the content of retention period. The content of the inter-trial interval was decoded as a baseline.
We observed one of the three events dominated the content of working memory. The category or the order of sequence did not determine what event would replay. Instead, the strength of attention at encoding was inversely related to delay-period decoding strength, suggesting the most weakly attended event at encoding was subsequently prioritized during the working memory interval.
Subsequent memory performance of the selectively replayed event was better than the memory of other events. Prioritizing the less-attended event at encoding in the maintenance is a resource-effective mechanism.
We propose the prioritization of weakly encoded events protects them from interference of other well-encoded events during the maintenance period. Whether strongly attended events were encoded into and retrieved from long-term memory or they were stored in an “activity silent” state remains an open question.
An intriguing question raised by our data is how the weakly encoded stimuli are prioritized for maintenance. Since prioritization was independent of sequence position, it could have only occurred after all three stimuli were encountered. A parsimonious scenario is that maintenance prioritization occurs at the beginning of the delay period and involves retrieval of information.
One possibility is the prioritized stimulus required more search or retrieval effort during the delay. Prefrontal mechanisms can support this mechanism by monitoring and controlling interference from strongly encoded stimuli. This potential mechanism would compensate for capacity limitations of working-memory.
In summary, we decoded the content of visual working memory with high temporal resolution using MEG. This technique allowed us to study replay dynamics for a sequence of events. The results revealed the representation of visual categorical information of the least attended stimulus during encoding was preferentially replayed during retention, and that this stimulus was better remembered than others.
Thus, working-memory selectively and intelligently amplifies the most weakly encoded memory event to maximize recall fidelity, instead of equally rehearsing the whole sequence. Our findings shift the functional role of working-memory from a faculty that “works with memory” to one that “works for memory” by actively selecting which encoded events need to be enhanced to be remembered.
Working Memory Replay Prioritizes Weakly Attended Events. Anna Jafarpour, Will Penny, Gareth Barnes, Robert T. Knight, Emrah Duzel. eNeuro Aug 2017, 4 (4) ENEURO.0171-17.2017; DOI: https://doi.org/10.1523/ENEURO.0171-17.2017